[BioC] DESeq and transcript-wise analysis

[Simon Anders]

Hi

On 2012-02-09 00:26, Elena Sorokin wrote:
> After re-reading related posts in the listserv archive, I still didn't
> know the exact answer to my question, so here goes. I'd like to use
> DESeq to measure differential isoform expression. Has Simon or anybody
> else written a script that will convert aligned reads (.bam/.sam file)
> into a table of isoform counts, suitable for input to DESEq - similar to
> what Simon has done at the gene-wise level, but instead for making a
> table of counts by isoform?

I get asked about every other week how to make such an isoform count 
table and, frankly, I still don't understand why anybody would want such 
a table.

A typical gene in a vertebrate genome has many isoform which all have 
large overlap with each other. Hence, if only one of the gene's several 
isoforms is differentially expressed, the extra reads from it will 
increase not only the count for this isoform but also for all other 
isoforms that have some overlap with it. Most likely, this means that 
_all_ isoforms have an increased count, because there is hardly ever two 
isoforms that do not overlap somehow. Hence, such a count table is not 
at all suitable to quantify isoform expression or to detect differential 
isoform usage.

I know this is obvious but I am so puzzled why so many people want to 
make such tables that I thought I should point it out again.

Quantifying isoform expression is a hard problem, and the various 
approaches discussed in the literature have a hard time to give 
quantifications which are precise enough to support tests for 
differential expression of isoforms. This is why we decided to take a 
whole step back and instead worked on a tool to test for differential 
usage of individual exons. We think that this approach, now available in 
the Bioconductor package "DEXSeq", is a more productive way to study 
alternative isoform regulation than looking at isoforms directly.

By the way, we now made a preprint of our manuscript on DEXSeq 
available: http://precedings.nature.com/documents/6837

   Simon